Super-resolution X-ray tomography using deep learning applied to the 3D quantification of defects in lattice...

Area of Science:

• Materials Science
• Additive Manufacturing
• Image Analysis

Background:

• Lattices are multiscale materials with defects impacting mechanical properties.
• X-ray computed tomography (CT) is vital for 3D defect characterization but faces resolution-time trade-offs.
• Current super-resolution validation lacks material science task-specific metrics.

Purpose of the Study:

• Redefine image quality for super-resolution from a material science perspective.
• Quantify measurement uncertainties in super-resolution for lattice defect characterization.
• Develop and validate a deep learning workflow for enhanced CT imaging of lattice defects.

Main Methods:

• Implemented a mixed-scale dense network for a comprehensive super-resolution workflow.
• Included data acquisition, preprocessing, and tailored algorithm validation.
• Experimentally tested on a laser powder bed fusion steel lattice.

Main Results:

• Super-resolution volumes accurately depicted porosity and surface roughness at various scales.
• Achieved significant reductions in scan time, spanning several orders of magnitude.
• Demonstrated improved morphometric depiction of defects.

Conclusions:

• Deep learning-based super-resolution enables feasible and efficient defect inspection in multiscale lattice materials.
• Task-based validation ensures the reliability of super-resolution for materials science applications.
• This approach overcomes CT limitations, making detailed defect analysis practical within reasonable timeframes.